by Sheldon Jackson
A number of methods have been proposed for predicting game winners in the National Collegiate Athletic Association’s (NCAA) annual men’s college basketball championship tournament. These methods utilize a variety of predictors, such as published rankings and season game results, and focus on the number of correctly predicted game outcomes. However, the analysis of these methods does not differentiate between the quality of predictions in different rounds of the tournament. Since 1985, more than 70% of the teams in the fourth, fifth, and sixth rounds of the tournament have been low-seeded teams (i.e., teams assigned seeds of one, two, or three by the NCAA selection committee); a method that can accurately compare two such teams is necessary to predict games in these rounds. While it is common to choose the team with the lower seed (i.e., with seed value closest to one) to win each game, this paper presents statistical hypothesis testing on tournament results from 1985 through 2008 that suggests there is an insignificant difference in the performance of low-seeded teams in the fourth, fifth, and six rounds of the tournament, while their performances differ in the first three rounds, where performance is defined as a seed’s historical win percentage in the round. This result suggests that winners in games between low-seeded teams in the fourth, fifth, and sixth rounds of the tournament cannot be accurately predicted using seeds alone, and alternate predictors should be sought.
Sheldon H. Jacobson is a Professor and Director of the Simulation and Optimization Laboratory at the University of Illinois at Urbana-Champaign. He has a B.Sc. and M.Sc. (both in Mathematics) from McGill University, and a Ph.D. (both in Operations Research and Industrial Engineering) from Cornell University. Douglas M. King is a Ph.D. student in the Department of Industrial and Enterprise Systems Engineering at the University of Illinois. He has a B.S. in General Engineering and a M.S. in Systems and Entrepreneurial Engineering, both from the University of Illinois. Their research interests focus on applying operations research methods to interesting problems of popular interest, such as the impact of obesity on fuel consumption, and optimal gerrymandering of congressional districts. The research in his poster grew out of their discussions on whether seeds in the NCAA “March Madness” are good predictors of performance across the six rounds of the tournament.